請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/7416
標題: | 完全卷積自動編碼器應用在心電圖訊號的壓縮與除噪 ECG Compression and Denoising based on Fully Convolutional Autoencoder |
作者: | Hsin-Tien Chiang 江欣恬 |
指導教授: | 簡韶逸(Shao-Yi Chien) |
關鍵字: | 訊號壓縮,除噪,心電圖訊號,自動編碼器,深度學習, Signal compression,Denoising,ECG signals,Autoencoders,Deep learning, |
出版年 : | 2018 |
學位: | 碩士 |
摘要: | 心電圖提供一種非侵入式檢測方法,廣泛且有效應用於檢測和預防心律不整的疾病像是早發性心室收縮。在長期心電監測而產生龐大資料量情況下,一個有效率的訊號壓縮處理方法是必要的。本論文提出一種利用完全卷積自動編碼器的心電圖壓縮方法。我們使用 MIT-BIH心律不整數據資料庫內的10筆數據來做壓縮效能的評估。與離散小波變換和深度自動編碼器相比,完全卷積網路在正常人和早發性心室收縮患者上都獲得了較佳的結果。此外,完全卷積網路相較於全連接神經網路在相同壓縮比下具有較小百分比的失真度。數據方面,在壓縮比為33.64,完全卷積網路對所有人達到平均失真度13.65%。由於心電訊號在量測過程中容易受到雜訊干擾,我們也驗證所提出的模型在去噪應用上的成效。對心電訊號加入不同信噪比的雜訊進行實驗,結果證明完全卷積網路具有比全連接神經網路更好的除噪效果。總結來說,完全卷積網路在壓縮方面產生較高的壓縮比和較低的失真度而除噪方面則有更高信噪比的提升,相信在臨床實踐中完全卷積網路具有良好的應用前景。 The electrocardiogram (ECG) is an efficient and non-invasive application for arrhythmia detection and prevention such as premature ventricular contractions (PVCs). Due to huge amounts of data generated by long-term ECG monitoring, an effective compression method is essential. This thesis proposed an autoencoder-based method utilizing fully convolutional network (FCN). The proposed approach is applied to 10 records of the MIT-BIH Arrhythmia database. Compared with discrete wavelet transform (DWT) and DNN-based autoencoder (AE), FCN acquires better performance for both normal and abnormal PVCs individuals. Moreover, FCN outperforms DNN in less percentage root-mean-squared difference (PRD) within identical compression ratio (CR). The proposed compression scheme exploits the fact that FCN achieves CR=33.64 with PRD=13.65% in average among all subjects. Because ECG signals are prone to be contaminated with noises during monitoring, we also validate FCN in denoising application. The results conducted on noisy ECG signals of different levels of signal-to-noise (SNR) show that FCN has better denoising performance than DNN. In summary, FCN yields higher CR and lower PRD in compression as well as higher SNR improvement in denoising and is believed to have good application prospect in clinical practice. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/7416 |
DOI: | 10.6342/NTU201804207 |
全文授權: | 同意授權(全球公開) |
電子全文公開日期: | 2028-10-12 |
顯示於系所單位: | 電子工程學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-107-1.pdf 此日期後於網路公開 2028-10-12 | 4.59 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。